Microsoft has recently officially launched a new-generation AI model focused on structured decision-making tasks — Microsoft-Decision-1. The model has now completed testing in multiple internal scenarios, demonstrating a huge advantage in running speed. In 36 authoritative benchmark tests covering approximately 150,000 questions, it ranked first in accuracy across the board, becoming a highly watched rising star in high-speed decision-making.
In terms of core functions and application scenarios, the model is primarily used to precisely select the optimal solution from preset options and can assign corresponding probability scores to each candidate answer. Its capabilities seamlessly cover multiple real-world business scenarios such as routing and content classification, and it can be easily embedded into existing applications, AI agents, and workflow systems, greatly improving the operational efficiency of complex workflows.
In terms of technical architecture and foundation, Microsoft-Decision-1 is built on Qwen3.5-9B and has undergone specialized post-training. Microsoft has officially revealed that it plans to migrate it to more other foundation models in the future. In terms of stability and safety, the model exhibits excellent robustness, with an extremely low decision flip rate after various perturbation treatments, while also achieving good results in multiple stringent safety tests.
In addition, the model is also highly competitive in terms of economic cost and commercial pricing. Its pricing standard is only $0.042 per million tokens on the input side, while the output side is completely free. Actual tests show that it can maintain a quality level comparable to some large models in similar decision-making tasks, but achieves a leap in running speed while greatly reducing usage costs.
Currently, the model has been officially opened to developers through Microsoft's AI development platform, and Microsoft also plans to further integrate it into the OpenRouter platform to provide more convenient invocation support for the developer ecosystem.